How to Use Logistic Regression with sklearn in Python

Use LogisticRegression from sklearn.linear_model by creating a model instance, fitting it with training data using fit(), and then predicting with predict(). This process trains the model to classify data based on input features.

Syntax

The basic syntax to use logistic regression in sklearn involves importing the class, creating an instance, fitting the model to data, and making predictions.

• LogisticRegression(): Creates the logistic regression model.
• fit(X_train, y_train): Trains the model on features X_train and labels y_train.
• predict(X_test): Predicts labels for new data X_test.

python

from sklearn.linear_model import LogisticRegression

model = LogisticRegression()
model.fit(X_train, y_train)
predictions = model.predict(X_test)

Example

This example shows how to train a logistic regression model on a simple dataset and predict the class labels.

python

from sklearn.linear_model import LogisticRegression
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score

# Load dataset
iris = load_iris()
X = iris.data
y = (iris.target == 0).astype(int)  # Binary classification: class 0 vs others

# Split data
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)

# Create and train model
model = LogisticRegression(max_iter=200)
model.fit(X_train, y_train)

# Predict
predictions = model.predict(X_test)

# Evaluate
accuracy = accuracy_score(y_test, predictions)
print(f"Accuracy: {accuracy:.2f}")

Output

Accuracy: 1.00

Common Pitfalls

Common mistakes when using logistic regression in sklearn include:

• Not scaling features when needed, which can slow convergence.
• Using default max_iter too low, causing the model not to converge.
• Confusing predict() (class labels) with predict_proba() (probabilities).
• Trying to use logistic regression for multi-class without specifying the right solver or multi-class option.

python

from sklearn.linear_model import LogisticRegression

# Wrong: default max_iter too low for some data
model = LogisticRegression()
model.fit(X_train, y_train)  # May warn about convergence

# Right: increase max_iter
model = LogisticRegression(max_iter=200)
model.fit(X_train, y_train)

Quick Reference

Method/Parameter	Description
LogisticRegression()	Create logistic regression model instance
fit(X, y)	Train model on features X and labels y
predict(X)	Predict class labels for X
predict_proba(X)	Predict class probabilities for X
max_iter	Maximum iterations for solver to converge (default 100)
solver	Algorithm to use (e.g., 'lbfgs', 'liblinear')
multi_class	'auto', 'ovr', or 'multinomial' for multi-class handling

Key Takeaways

Import LogisticRegression from sklearn.linear_model to create the model.

Always fit the model with training data before predicting.

Increase max_iter if the model does not converge.

Use predict() for class labels and predict_proba() for probabilities.

Scale features if convergence is slow or data varies widely.